Issue Number: Vol. 1, No. 1
Year of Publication: 2011
Page Numbers: 195-203
Authors: Ismail Burak PARLAK, Salih Murat EGI, Ahmet ADEMOGLU, Costantino BALESTRA, Peter GERMONPRE, Alessandro MARRONI
Journal Name: International Journal of Digital Information and Wireless Communications (IJDIWC)
- Hong Kong


In cardiology, automatic recognition and image analysis still conserve computational challenges in terms of medical guidance and diagnosis. Bubbles or microemboli that circulate into cardiopulmonary system are considered suspicious for several medical problems such as decompression sickness, stroke and migraine. The aim of our work is to develop and assess an automatic approach to detect these bubbles that are analyzed manually by clinicians. In this paper, five different divers were examined in post decompression phase and their cardiac videos were recorded. Existent bubbles on the frames were also marked by two specialists in order to compare with our results. We developed our neural network architecture by integrating Gabor-Wavelet kernel which is commonly used in face and pedestrian recognition. The training phase of the network was performed using real bubble morphologies. Our recognition phase was achieved on four cardiac chambers through echocardiographic frames. Our correct detection ratio was between 77.6- 94.3%. We consider that our approach would be useful in longitudinal researches in hypobaric and hyperbaric environments and risky subjects with congenital defects.